The day in one read

The NVIDIA Scandal and AI Valuation Correction

The dominant story of the day is the intersection of a corporate governance crisis at NVIDIA and a broader, sharp correction in artificial intelligence valuations. NVIDIA CEO Jensen Huang admitted in a company-wide email to an affair with Alice Huang, an executive at Singaporean cloud provider Megaspeed, whom U.S. intelligence suspects is an agent for China’s Ministry of State Security. The NVIDIA board launched an independent investigation, stating it has no reason to believe national security or company secrets were compromised, but the revelation threatens Huang’s influence in advocating for normalized U.S.-China AI trade relations. Megaspeed has provided Chinese AI labs with remote access to banned NVIDIA GPUs and is under Commerce Department investigation for smuggling chips.

Market reaction was volatile, with NVIDIA stock falling 8.8% intraday before recovering half the loss, and S&P 500 futures initially erasing $300 billion in market cap. The broader AI sector has lost between $1.4 and $1.5 trillion in value since October, with the semiconductor index valuation dropping from 33x to 14x expected earnings. This correction is reflected in private markets as well; Anthropic’s IPO valuation estimates have collapsed from $2 trillion to as low as $1.1 trillion, with the company aiming to raise $80 billion at a $1.5 trillion valuation. OpenAI closed a funding round at $1.25 trillion, securing $11 billion against an initial $30 billion target. Analysts debate whether Huang is "indispensable" to U.S. AI leadership, while some argue the scandal serves to force through critical objectives in the tech war that Huang had previously opposed.

AI Infrastructure: Centralization, Decentralization, and Sovereignty

Strategic disagreements are emerging over the optimal architecture for AI infrastructure, pitting centralized hyperscaler models against decentralized and sovereign alternatives. Pete Sacco, founder of Gray Wolf, argues that the industry is shifting from centralized training to decentralized inference. He criticizes the "Neocloud" model, exemplified by CoreWeave and Lambda, for signing 20-year leases on co-location space and financing GPUs with only a two-to-three-year lifespan, creating significant financial risk. Sacco projects that the next major AI market is inferencing, which does not require 100% GPU training stacks, and aims to build a "data center in every town" to provide local sovereignty.

In Europe, Mistral CEO Arthur Mensch positions sovereign AI as a structural economic imperative, arguing that reliance on U.S. or Chinese providers creates trade balance deficits and security risks. Mistral, valued at $21 billion, is vertically integrated, managing supply chains for GPUs and hosting energy-intensive inference in owned infrastructure. Mensch asserts that energy constitutes only 10% of total compute costs, with chip and capital costs dominating, and predicts the inference market will behave like a commodity. Meanwhile, Red Hat’s Eduardo Vache detailed the LLMD project, a distributed inference framework for Kubernetes that complements the vLLM engine. LLMD manages routing and scheduling across multiple instances to maintain stateful KV caches, aiming to maximize GPU value and enable private, on-premises AI by reducing reliance on expensive frontier models for routine tasks.

Frontier Model Capabilities and Data Bottlenecks

Anthropic released Fable 5.2, a frontier model positioned for scope, complexity, and creativity, alongside Claude Mythos 5.2, a variant restricted to cybersecurity and life sciences. Fable 5.2 outperforms other frontier models on benchmarks including Frontier Code and Automation Bench, demonstrating significant speed in complex tasks like animating sprites and 3D rendering. However, the model exhibits poor taste in front-end web design and lacks native image generation capabilities, often requiring pairing with OpenAI’s image gen 2.5. While Anthropic states that most daily coding tasks remain suited for Opus 5.5, Fable 5.2 is highlighted for solving two open problems in black hole physics.

Despite these advancements, Handshake AI argues that high-quality, expert-authored data is the critical bottleneck for AI progress. The frontier data market has expanded from $1 billion three years ago to $10 billion currently, with a forecast of $100 billion in three years. Handshake’s benchmarks reveal that frontier models are functionally unusable in high-stakes domains; in life sciences, models perform at 50% or worse on lab assay interpretation, and in finance, they score sub-20% on complex tasks like LBO modeling. Experts note that if output is 95% right, it is 100% wrong because fixing flawed outputs takes longer than rebuilding from scratch. This suggests that while compute and model size continue to scale, alignment and data quality are now the primary drivers of practical utility.

The Rise of Agentic Workflows and Tokenized Finance

AI agents are increasingly automating complex operational tasks, from marketing to software development, while tokenization is reshaping financial markets. Cody Schneider argues that marketing is transitioning from human labor to code and tokens, enabling AI agents to function as "marketing engineers." He outlines a workflow using platforms like Graph and FAL to generate static ads and UGC avatar videos at a cost of approximately $5 per ad, compared to $500 for traditional production. Similarly, William Emo, founder of Hackmamba, runs a 25-person company as a solo founder using an "army" of AI agents for all coding and operational tasks. Emo built a native iOS and Android application for learning Norwegian using Codex and GPT Live, emphasizing the importance of design language and product specificity in prompting.

In finance, tokenized stock trading is accelerating. On November 9, $657 million in tokenized stocks traded on Solana, with 90% of volume on Backpack. Fidelity reported that Solana briefly traded more stocks than the New York Stock Exchange. Backpack CEO Armani Ferrante describes the market as entering a "token super cycle," positioning the platform as a "neo brokerage" that allows users to buy tokenized stocks in DeFi and redeem them into traditional brokerage accounts. Ferrante argues that tokenization is an inevitable, generational shift in finance that will bring the entire global economy on-chain, with the addressable market encompassing the entire economy.

Also Notable

US midterm elections triggered significant market volatility, with the US semiconductor index dropping over 10 percent and the Nasdaq losing approximately $2.5 trillion in market value. OpenAI announced new rules for socially compatible datacenters, stating that residents should not pay for datacenter electricity consumption and that environmental impact data would be published. In Germany, a study by the Öko-Institut projects AI datacenter water consumption could double by 2030, while the EU is investing over 500 million euros in a chip-design platform to boost European capacity.

Berkshire Hathaway reported Q3 operating earnings down 5.8% to $12.7 billion, with insurance profits falling 43%. CEO Greg Abel repurchased $9.6 billion in shares, leaving a cash pile of $377 billion. Airbnb reported 19% revenue growth, its strongest since 2023, with CEO Brian Chesky attributing this to AI-accelerated product testing. The stock rose over 10% to a valuation of approximately $100 billion. In China, robotics stocks are down 41.8% over five years despite the country installing 295,000 industrial robots in 2024, representing 54% of the global total. A new HSBC MSCI China Robotics Usage ETF launched on Xetra, tracking 38 companies with at least 50% robotics revenue.